Re: [PATCH] libertas: remove internal buffers from GSPI driver

On Mon, 2009-11-30 at 16:53 -0800, Colin McCabe wrote:
> On Tue, Oct 27, 2009 at 3:51 PM, Andrey Yurovsky <andrey@...> wrote:
> > This patch removes the internal command and data buffers that the GSPI driver
> > maintained and instead relies on the Libertas core to synchronize access
> > to the command and data ports as with the other interface drivers.
> 
> Hi all,
> 
> I'm glad to see you're improving stuff. Good catch with some of the
> recent bugfixes.
> 
> However... it seems like you are headed for a "scheduling while
> atomic" with this patch.
> 
> The original reason why if_spi_host_to_card copied the data into a
> buffer, rather than simply calling spu_write and being done with it,
> was that spu_write needed to sleep. spu_write invokes spi_sync, which
> has a big comment that says "This call may only be used from a context
> that may sleep."
> 
> Unfortunately, if_spi_host_to_card is called from contexts where we
> hold a spinlock. For example, lbs_main does
> 
> spin_lock_irq(&priv->driver_lock);
> if (!priv->dnld_sent && priv->tx_pending_len > 0) {
>     int ret = priv->hw_host_to_card(priv, MVMS_DAT,
> 
> priv->tx_pending_buf,
> 
> priv->tx_pending_len);